.jpg)
Calling Freebuff AI just a "coding assistant" is an understatement. It's actually a multi-agent system with 9 specialized sub-agents — each an expert in its own domain. In this article, we'll deep dive into every agent: what it does, when to use it, how to invoke it, and pro tips that will 10x your productivity.
1. What is Multi-Agent Architecture?
Traditional AI coding tools (like Claude Code, Copilot) use a single model that handles everything — from file finding to code writing, testing, and review. The problem? A single model is average at everything and exceptional at nothing.
Freebuff has completely reversed this approach. It's an orchestrator that analyzes your task and deploys the right agent at the right time. Need to find files? Gemini Flash Lite (ultra-fast). Need to write code? DeepSeek V4 Pro (high-quality). Browser test? Dedicated Browser agent. This approach delivers both parallelism and specialization.
2. Agent #1: Coding Agent (Primary)
This is Freebuff's primary agent and handles most of your interactions. Whenever you describe a coding task — whether creating new files, editing existing code, or performing multi-file refactoring — this agent takes charge.
Capabilities:
- Create new files (HTML, CSS, JS, Python, PHP, etc.)
- Make precise edits to existing files
- Multi-file changes in a single conversation
- Import management and dependency resolution
- Error handling and edge case coverage
How to invoke: It's the default agent — just describe your task directly. Use @filename to target a specific file.
> Create a responsive navigation component with hamburger menu for mobile, dark theme, and smooth animations
✓ Created: components/Navbar.jsx
✓ Created: styles/navbar.css
✓ Updated: App.jsx (import added)
📎 Follow-ups:
[1] Add dropdown menus to navigation
[2] Add active link highlighting
[3] Test on mobile viewport
3. Agent #2: File Picker (Gemini-Powered)
This is Freebuff's speed secret weapon. Whenever the coding agent needs files, the File Picker activates first and identifies relevant files in seconds — even in codebases with thousands of files. Gemini 3.1 Flash Lite is specifically optimized for ultra-fast file scanning.
How it works:
- Scans the codebase file tree
- Analyzes file names, paths, and partial content
- Selects the most relevant files based on the task
- Passes only focused context to the Coding Agent (faster processing)
This is the reason Freebuff is 5-10x faster than Claude Code — context gathering happens in parallel, using a fast specialized model.
@filename to point to a specific file when you want to give the File Picker a hint. For example: "@utils/auth.js fix the JWT expiry issue" — this makes the File Picker grab that file directly without scanning, resulting in even faster output.4. Agent #3: Code Reviewer
This agent gives your code a harsh but fair review. It highlights bugs, security vulnerabilities, performance issues, and best practice violations. Invoke it using the /review command or by mentioning @reviewer.
It reviews for:
- Security: SQL injection, XSS, CSRF vulnerabilities
- Performance: Memory leaks, unnecessary re-renders, slow queries
- Best Practices: Naming conventions, code duplication, SOLID principles
- Edge Cases: Null checks, error handling, boundary conditions
> /review
🔍 Reviewing recent changes...
⚠ SECURITY: auth.js:42 — User input directly in SQL query (SQL injection risk)
⚠ PERFORMANCE: dashboard.jsx:108 — useEffect missing dependency array (infinite re-render)
✓ GOOD: Proper error handling in api/routes.js
✓ GOOD: Consistent naming conventions
Overall: 7/10 — 2 critical issues found. Fix recommended.
[1] Fix SQL injection vulnerability
[2] Fix useEffect dependency
[3] Add input validation layer
5. Agent #4: Browser Use Agent
This is Freebuff's most unique feature — one that doesn't exist in Claude Code, Copilot, or any other competitor. The Browser Use agent controls a real browser (Chromium-based), takes screenshots, inspects the DOM, and checks CSS variables.
Capabilities:
- Visual Testing: Loads the page and performs screenshot-based analysis
- Responsive Testing: Tests across multiple viewport sizes (mobile, tablet, desktop)
- DevTools Integration: CSS computed styles, console errors, network requests
- Navigation: Clicks links, fills forms, and simulates user flows
- Comparison: Before/after screenshots for UI changes
> Test my contact form — fill in dummy data, submit it, and check if success message appears
🌐 Opening localhost:3000/contact...
📝 Filling form: Name="Test User", Email="test@example.com", Message="Hello"
🖱️ Clicking "Submit" button...
📸 Screenshot after submit captured.
✓ Success: "Thank you!" message appeared in .success-toast
⚠ Note: Form does not reset after successful submit
⚠ Note: No loading spinner during API call (UX improvement)
[1] Add form reset after success
[2] Add loading spinner to submit button
[3] Test with invalid email format
6. Agent #5: Deep Thinking (GPT-5.4)
This agent is built for complex reasoning tasks — when simple code generation isn't enough. It activates for architecture decisions, algorithm design, system planning, and multi-service integration tasks.
How to activate: Link your ChatGPT Plus subscription using the /connect command. After that, whenever Freebuff encounters a complex task, GPT-5.4 is automatically invoked.
Important Note: This is the only agent that isn't completely free — it requires a ChatGPT Plus ($20/month) subscription. The other 8 agents are entirely free. However, if you already have a ChatGPT subscription, this is an incredible bonus.
7. Agents #6-9: Research, Bash, Performance, Interview
When Freebuff needs answers about an unfamiliar API, library, or error, the Web Research agent performs a live internet search. It fetches the latest documentation, Stack Overflow answers, and GitHub issues in real-time. This is especially helpful when you're working with new frameworks or libraries.
Run terminal commands directly from Freebuff — without switching windows. Use the ! prefix or /bash mode. npm install, git operations, build scripts, docker commands — everything executes from within Freebuff.
> !npm install express cors helmet
added 3 packages in 4s
✓ Dependencies installed.
> !git status
On branch main
Changes not staged for commit:
modified: src/server.js
modified: package.json
Is your website slow? The Performance agent runs a Lighthouse-style analysis and provides actionable optimization recommendations. Image optimization, code splitting, lazy loading, caching strategies — it suggests everything and can even implement the fixes if you agree.
Activated via the /interview command. When you have a vague idea and can't clearly describe it, the Interview agent asks smart questions to clarify your requirements. Based on your answers, it generates a clear spec, and then the Coding Agent implements that spec. This feature prevents wrong code generation that typically results from vague prompts.
> /interview
📋 Interview Mode — I'll ask questions to understand your requirements.
Q1: What type of project is this? (website, API, CLI tool, mobile app?)
> website
Q2: What's the main purpose? (landing page, e-commerce, blog, dashboard?)
> portfolio website
Q3: Any specific tech stack preference?
> React + Tailwind
Q4: How many pages do you need?
> home, about, projects, contact
✓ Spec generated! Starting implementation...
8. Advanced Workflows: Chaining Agents
Freebuff's real power emerges when you chain agents together — describe a task and let multiple agents automatically collaborate. Below are some powerful workflow examples:
Workflow 1: Build → Test → Fix
Workflow 2: Research → Plan → Implement
9. knowledge.md: Custom Project Intelligence
The knowledge.md file serves as Freebuff's persistent memory that is read at the start of every conversation. In this file, you can document your project's tech stack, naming conventions, architectural decisions, and important notes. Better knowledge = better code generation.
# Project: E-Commerce Platform
## Tech Stack
- Frontend: Next.js 15, TypeScript, Tailwind CSS
- Backend: Express.js, PostgreSQL, Redis
- Auth: NextAuth.js v5
## Conventions
- Components: PascalCase (UserCard.tsx)
- Utils: camelCase (formatPrice.ts)
- API routes: kebab-case (/api/user-orders)
## Important Rules
- Always use server components by default
- Client components only when useState/useEffect needed
- All API responses: { success, data?, error? }
- Use Zod for input validation on all API routes
## Current Sprint
- Implementing checkout flow
- Stripe integration pending
- Mobile responsiveness 80% complete
10. Pro Tips & Best Practices
@filename to point to specific files. Use @agent to invoke a specific agent. This dramatically improves context quality and prevents irrelevant changes./interview first. The AI will ask smart questions and generate a clear spec. Then implement. This is 10x better than coding in the wrong direction./init as soon as you start a project. Clearly document your tech stack, conventions, and rules. This file is Freebuff's "memory" — better memory = better code.
